TODOs for next build

[h-vetinari]

Since rebuilding takes so much effort and we don't do it that often, I wanted to create an issue to track the various things we should adapt for the next build:

EDIT: "checked" below means it's currently included in #176

• [x] bump to tensorflow 2.7.0
  • [x] needs keras-builds: https://github.com/conda-forge/keras-feedstock/pull/56
  • [ ] needs packaging tensorflow-io-gcs-filesystem: PyPI github
• [ ] remove bazel-workarounds that should now be taken care of in bazel-toolchain (see https://github.com/conda-forge/bazel-toolchain-feedstock/pull/5)
• [x] next gprc migration:
  • https://github.com/conda-forge/tensorflow-feedstock/pull/177
• [x] Add cudatoolkit as a dependency to tensorflow, or even better, get rid of tensorflow-base (since it's not needed anymore now that tf-estimator is built in this feedstock), see https://github.com/conda-forge/tensorflow-feedstock/issues/162
  • note that "getting rid" means in terms of the core recipe, there could/should still be a compatibility output for existing users in that case.
• [x] Loosen package pinnings where possible, see #151
  • [x] In particular, numpy has been unpinned from 1.19.2 in TF 2.7 as of https://github.com/tensorflow/tensorflow/commit/a90383a867a2337600c775d67842f308da9d118b
• [x] Add a run-depend on __cuda virtual package or the GPU builds, see #174
• [x] Python 3.10 builds? The migrator is currently failing to resolve the builds, probably not least due to the tight package pinnings (and there being no numpy 1.19 for cpython 3.10).
• [x] Remove gilbertfrancois from the maintainers: https://github.com/conda-forge/tensorflow-feedstock/issues/178
• [ ] Unskip libtensorflow{,_cc}, see #181.

CC @hmaarrfk @xhochy @wolfv @conda-forge/tensorflow, feel free to add more things in the description. If any of those points end up being a huge hassle, it can of course be pushed off to the build after that.

Anyone is welcome to chime in if I forgot something. 🙃

[njzjz]

Currently tensorflow here is not built against MKL (OneDNN), but do we plan to?

[h-vetinari]

Currently tensorflow here is not built against MKL (OneDNN), but do we plan to?

Not sure how involved this is, but since onednn seems to be available for all relevant platforms, I'm not opposed to trying!

[xhochy]

Tensorflow is building oneDNN from source in the build currently. That is probably the first 1-2h of CI. Getting that from the feedstock instead of building from source would help a lot.

[h-vetinari]

Tensorflow is building oneDNN from source in the build currently. That is probably the first 1-2h of CI. Getting that from the feedstock instead of building from source would help a lot.

That sounds like a big win if we can pull it off!